Causal effect estimation method, system and equipment of electric power energy consumption control strategy without observation confusion and medium

Abstract

The invention belongs to the technical field of causal inference, and discloses a causal effect estimation method, system, equipment and medium for a power energy consumption control strategy under unobserved confusion, so as to solve the problem that unobserved confusion variable interference cannot be eliminated. The method comprises the following steps: acquiring a training data set; end-to-end joint iterative training is carried out on the long sequence modeling module and the anti-fact regression module, and the joint iterative training comprises the steps that electricity consumption historical information, an energy consumption management and control strategy sequence and an electricity consumption sequence at historical moments are input into the long sequence modeling module, and confusion variable representation and tool variable representation are obtained through learning; inputting the confusion variable representation, the energy consumption management and control strategy sequence, the tool variable representation and the predicted energy consumption management and control strategy at the next moment into an anti-fact regression module, and predicting the electricity consumption at the next moment; and inputting to-be-estimated sequence data into the trained long sequence modeling module and the anti-fact regression module, and predicting to obtain the power consumption under the target energy consumption control strategy.

Description

Technical Field

[0001] This invention belongs to the field of causal inference technology, and specifically relates to a method, system, device and medium for estimating the causal effects of power consumption management strategies under unobserved confusion. Background Technology

[0002] Estimating the causal effects of sequential interventions is a core challenge in the refined management of power energy. Power supply companies need to formulate and adjust energy management strategies (intervention variables) based on dynamic information such as historical electricity load, weather, and holidays to achieve precise control over regional electricity consumption (outcome variable). However, this decision-making process is severely hampered by unobserved confounding variables, such as sudden population movements, the implicit aging of power equipment, and fluctuations in energy-saving awareness among users. These factors not only affect strategy formulation but also directly impact electricity consumption outcomes, making it difficult for traditional methods to accurately estimate the true effectiveness of management strategies.

[0003] Currently, causal inference methods based on observational data heavily rely on the "no-confounding assumption," which assumes that all common influencing factors have been measured. This premise is difficult to hold in complex power systems. Although some cutting-edge studies have attempted to address this by using time series models (such as Markov models or autoregressive models) or introducing instrumental variables, they have inherent limitations: time series models are computationally inefficient when dealing with high-dimensional long series and still assume that confounding variables can be fully observed; while identifying effective instrumental variables (such as specific policies) strongly relies on expert prior knowledge and lacks universality.

[0004] In summary, existing technologies have fundamental limitations in addressing the problem of causal inference in power systems: first, they cannot automatically learn effective instrumental and confounding variable representations from historical data; second, they fail to effectively eliminate the interference of unobserved confounding variables on the estimation of the causal effects of serialized energy consumption management strategies. Summary of the Invention

[0005] Based on the aforementioned shortcomings and deficiencies in the prior art, one of the objectives of this invention is to at least solve one or more of the aforementioned problems in the prior art. In other words, one of the objectives of this invention is to provide a method, system, device, and medium for estimating the causal effects of power consumption management strategies under unobserved confusion that meets one or more of the aforementioned requirements. This method automatically eliminates unobserved confusion through an end-to-end framework, thereby achieving accurate estimation of the causal effects of power sequence interventions.

[0006] To achieve the above-mentioned objectives, the present invention adopts the following technical solution: In a first aspect, the present invention provides a method for estimating the causal effects of power consumption management strategies under unobserved confusion, comprising the following steps: S1. Obtain the training dataset, which includes historical electricity consumption information at historical moments. Energy consumption management strategy sequence and electricity consumption sequence ; S2. Based on the training dataset, perform end-to-end joint iterative training on the long sequence modeling module and the counterfactual regression module, wherein the joint iterative training includes: S21. The electricity consumption history information at the aforementioned historical time point... Energy consumption management strategy sequence and electricity consumption sequence The input is fed into the long sequence modeling module to learn the representation of the confusion variables. and instrumental variable representation The confusion variable characterizes The instrumental variables are used to characterize incompletely observed variables that simultaneously affect energy management strategies and electricity consumption. Used to characterize variables that are relevant to energy management strategies and are not subject to unobserved confounding; S22, Characterize the confusion variables The energy consumption management strategy sequence The instrumental variables represent And the energy management strategy expected in the next moment. The data is input into the counterfactual regression module to predict the electricity consumption at the next moment. ; S3. Input the sequence data to be estimated into the long sequence modeling module and the counterfactual regression module after training, execute steps S21 and S22, and predict the electricity consumption under the target energy control strategy to complete the causal effect estimation.

[0007] As a preferred embodiment, the long sequence modeling module includes a basic temporal representation learning network. Confusion variable representation network and instrumental variable representation network Step S21 is as follows: The electricity consumption history information at the aforementioned historical moments Energy consumption management strategy sequence and electricity consumption sequence Combined into a historical representation sequence ; Using the aforementioned basic temporal representation learning network For the historical representation sequence Processing is performed to obtain the basic representation. ; The basic representation Input the confusion variable characterization network To obtain the characterization of the confounding variables ; The basic representation Input the instrumental variable characterization network To obtain the instrumental variable representation .

[0008] As a preferred embodiment, the counterfactual regression module includes an outcome prediction network. and energy consumption management strategy prediction network Step S22 is as follows: Characterize the confounding variables The energy consumption management strategy sequence and the energy management strategy for the next expected moment. Input the result prediction network The electricity consumption forecast results were obtained. ; Characterize the confounding variables The energy consumption management strategy sequence and the instrumental variable characterization Input the policy prediction network The prediction results of the energy consumption management strategy are obtained. .

[0009] As a preferred approach, in the joint iterative training of step S2, the total loss function of the optimization algorithm is optimized. To update the model parameters, the total loss function The expression is: , In the formula, For electricity consumption prediction results based on the next time point and actual electricity consumption results The predicted loss of the resulting variable is calculated. The mutual information loss for overall representation learning, To counteract the generalized moment conditional loss, and This is a hyperparameter.

[0010] As a preferred approach, during the training process of the long sequence modeling module, the mutual information loss of the overall representation learning is calculated. Its functional form is: , In the formula, For instrumental variable representation And the energy management strategy for the next moment. The calculated correlation mutual information loss of the instrumental variables, For instrumental variable representation And the actual power consumption results at the next moment. And characterized by confounding variables Energy consumption management strategy sequence Energy management strategy for the next moment and electricity consumption sequence The exclusive mutual information loss of the instrumental variables obtained by conditional calculation. For the representation of confounding variables And the energy management strategy for the next moment. The calculated mutual information loss of the correlation between the confounding variable and the intervention variable, For the representation of confounding variables And the actual power consumption results at the next moment. The calculated mutual information loss of the correlation between the confounding variable and the outcome variable, For instrumental variable representation and confounding variable representation The calculated mutual information loss of independence between the instrumental variable and the confounding variable.

[0011] As a preferred embodiment, the functional form of the adversarial generalized moment conditional loss is: , In the formula, For the sample size, The length of the time series. To predict the network output for energy management strategies, The output of the prediction network is given the result. This represents the actual electricity consumption.

[0012] As a preferred option: The basic temporal representation learning network The result prediction network All use the transformer model; The confusion variable characterization network The instrumental variable representation network and the energy consumption management strategy prediction network All models employ a multilayer perceptron model.

[0013] Secondly, the present invention provides a causal effect estimation system for power consumption management strategies under unobserved confusion, for implementing the causal effect estimation method as described in the first aspect.

[0014] Thirdly, the present invention provides an electronic device, the computer device including a memory, a processor and a computer program, wherein when the computer program is executed by the processor, it implements the causal effect estimation method as described in the first aspect.

[0015] Fourthly, the present invention provides a computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, implements the causal effect estimation method as described in the first aspect.

[0016] Compared with the prior art, the present invention has the following beneficial effects: 1. Effectively overcomes estimation bias caused by unobserved confounding. This invention constructs an end-to-end learning framework that integrates double negative control variables and generalized moment conditions, enabling accurate estimation of the causal effects of sequential interventions even in the presence of unobserved confounding, thus overcoming the strong dependence of traditional methods on the "no confounding hypothesis".

[0017] 2. Automating variable learning reduces reliance on expert knowledge. Unlike previous methods that required manual identification of instrumental variables, this invention proposes a data-driven approach to automatically learn effective instrumental and confounding variable representations from historical time-series data, improving the applicability and generalizability of the method in real-world scenarios.

[0018] 3. Learning ability that balances temporal dependencies and confounding structures. Through joint training of the long sequence modeling module and the counterfactual regression module, this invention simultaneously captures complex temporal dependencies and causal structures between variables, solving the common problems of insufficient representation ability of traditional time series models in high-dimensional long sequence scenarios and difficulty in handling unobserved confounding.

[0019] 4. Provides reliable basis for dynamic decision-making in power systems. This invention can accurately assess the causal impact of different energy consumption control strategies on electricity consumption, providing stable and reliable data support for dynamic decision-making tasks such as load forecasting and strategy optimization in power systems.

[0020] Further or more detailed beneficial effects will be described in conjunction with specific embodiments in the detailed implementation. Attached Figure Description

[0021] To more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0022] Figure 1This is a schematic diagram illustrating the application of the causal effect estimation method described in this invention in the medical field.

[0023] Figure 2 This is a schematic diagram illustrating the general causal relationships between various variables in the causal effect estimation method described in this invention.

[0024] Figure 3 This is a schematic diagram illustrating the principle of the causal effect estimation method described in this invention.

[0025] Figure 4 This is a schematic diagram of the joint iterative training process of the causal effect estimation method described in this invention.

[0026] Figure 5 This is a structural diagram of the electronic device described in Embodiment 3 of the present invention.

[0027] Figure 6 This is a schematic diagram of the visualization results described in Embodiment 5 of the present invention.

[0028] Icon labels: 500. Electronic devices; 501. Processor; 502. Communication bus; 503. User interface; 504. Network interface; 505. Memory. Detailed Implementation

[0029] The technical solutions of the present invention will be clearly and completely described below with reference to the accompanying drawings in the embodiments of the present invention.

[0030] In the following description, several embodiments of the present invention are provided. Different embodiments can be substituted or combined. Therefore, the present invention can also be considered to include all possible combinations of the same and / or different embodiments described. Thus, if one embodiment includes features A, B, and C, and another embodiment includes features B and D, then the present invention should also be considered to include embodiments containing one or more other possible combinations of A, B, C, and D, even if such embodiments are not explicitly described in the following text.

[0031] The following description provides examples and does not limit the scope, applicability, or examples set forth in the claims. Changes may be made to the function and arrangement of the described elements without departing from the scope of the invention. Various processes or components may be appropriately omitted, substituted, or added to the various examples. For example, the described methods may be performed in a different order than described, and various steps may be added, omitted, or combined. Furthermore, features described with respect to some examples may be combined into other examples.

[0032] To facilitate a better understanding of the embodiments of the present invention, its application scenarios will be explained before providing a detailed explanation of the specific implementation methods.

[0033] The causal effect estimation method described in the embodiments of this specification is applied to process power energy management and decision support. Its core lies in accurately assessing the causal impact of serialized energy consumption control strategies on electricity consumption, providing data-driven basis for grid dispatch and optimization.

[0034] Please see Figure 1 , Figure 1 The diagram illustrates the application of the causal effect estimation method in a medical context. It also shows a potential application example of the general technical framework described in this invention in another field, illustrating the universality of the method in addressing the problem of estimating the causal effects of interventions in sequences with unobserved confounding.

[0035] like Figure 1 As shown in this medical case, the observable characteristics of a patient with gastric cancer (corresponding to electricity consumption history in the power sector) may include: age, gender, income, body mass index, medical history, examination indicators, etc. Their unobservable characteristics (corresponding to unobserved confounding in the power sector) include: socioeconomic status, lifestyle, environmental exposure, mental health, drug metabolism, etc. By applying optional interventions (such as different combinations of chemotherapy or radiotherapy) to the patient at different times, corresponding to energy management strategies, the corresponding tumor volume (corresponding to electricity consumption) can be predicted, thereby dynamically determining the intervention plan for future moments.

[0036] Figure 2 This is a schematic diagram illustrating the general causal relationships between the variables in this invention. As can be seen from the simplified causal diagram, for intervention at the next time step... And the corresponding future results of interest All historical interventions and confounding variable representation This constitutes a complete set of observable confounding variables. (Result from the previous time step) It can serve as a natural negative control outcome (NCO) variable, while the instrumental variable representation decoupled from the long sequence modeling module can be regarded as a special case of negative control exposure (NCE) variable.

[0037] It should be noted that the above medical cases are merely examples to demonstrate that the technical framework proposed in this invention can be universally applied to different technical fields with similar causal structures to the power industry (i.e., the presence of time dependence, sequential intervention, and unobserved confusion). The scope of protection of this invention is not limited to this specific medical application, nor should it be construed as a limitation on the diagnosis or treatment of diseases. Its core protection content is still based on the technical solutions in the field of power energy management as clearly defined in the claims.

[0038] In these scenarios, the application of the causal effect estimation method aims to verify the applicability and effectiveness of the algorithmic framework for causal inference tasks with similar intermediate problem structures across different domains (i.e., the presence of temporal dependencies, sequential interventions, and unobserved confusion). For example, in a medical setting, the patient's treatment plan sequence can be used as an intervention variable, and disease indicators such as tumor volume as outcome variables. Similarly, this method can be used to estimate the causal effects of different treatment pathways, assisting in medical decision-making. It should be noted that such cross-domain application examples are only used to illustrate the generality of the invention, and the scope of protection of the invention is still limited to its application in the power industry as explicitly defined in the claims.

[0039] The following is a brief explanation of the unobserved confusion, power consumption management strategy, causal effect estimation, end-to-end joint iterative training, confusion variable representation, and instrumental variable representation involved in several embodiments of this specification: Unobserved confounding refers to variables in the power system that influence both the formulation of energy management strategies and final electricity consumption, but are difficult to observe directly and comprehensively. Examples include sudden regional population movements, fluctuations in energy-saving awareness among user groups, or the latent aging of power equipment.

[0040] Electricity consumption management and control strategies refer to a series of measures formulated and implemented by power supply companies to regulate electricity load and ensure grid stability, such as dynamic electricity pricing, load control instructions, and demand-side response incentives.

[0041] In this invention, causal effect estimation specifically refers to the process of quantitatively assessing the causal relationship between different "series of energy management strategies" and "future electricity consumption". Its goal is to accurately answer the counterfactual question of "how electricity consumption will change if another set of management strategies is adopted".

[0042] End-to-end joint iterative training refers to placing the long sequence modeling module and the counterfactual regression module in a unified deep learning framework, calculating the prediction results through forward propagation, and simultaneously optimizing the parameters of all modules through backpropagation, ultimately forming a complete and collaborative learning system that goes directly from the raw data to the estimation of causal effects.

[0043] The confounding variable representation refers to a low-dimensional vector learned from observable historical power data through the long-sequence modeling module, used to approximate the potential impact of unobserved confounding variables.

[0044] Instrumental variable representation refers to variables learned from observable historical power data through the long-sequence modeling module that are related to energy management strategies but not to unobserved confounding. As a data-driven "tool", it is used to effectively eliminate confounding bias in causal estimation.

[0045] Example 1: This embodiment provides a method for estimating the causal effects of power consumption management strategies under unobserved confusion, including the following steps: Step 1: Given a time Historical information on various electricity consumption (Information such as holidays, traffic, electricity consumption patterns, economic activity levels, weather, electricity prices, energy-saving subsidies, etc.) Energy consumption management strategies and historical electricity consumption transmission Entering the long sequence modeling module, it learns the confounding variables. (Simultaneously affecting energy management strategies and electricity consumption, such as holidays, traffic, electricity-consuming entities, economic activity levels, weather, etc.) and instrumental variables. (Characteristics related to energy management strategies, but not confounded by unobserved factors, such as electricity price adjustment policies and energy-saving subsidy policies). Figure 3 The lower left half shows the processing flow of the long sequence modeling module.

[0046] It should be noted that the long sequence modeling module in this embodiment includes a basic time series representation learning network. Confusion variable representation network Instrumental variable representation network In the long sequence modeling module, the first... During each training round, the parameters of the long sequence modeling module, updated in the previous training round, are obtained. Historical electricity consumption information Energy consumption management strategy sequence at historical moments and historical electricity consumption sequences As input The basic temporal representation learning network is used To obtain the basic representation ; to basic representation Input to the confusion variable representation network In this process, the characterization of the confounding variables is obtained. The basic representations are input into the instrumental variable representation network. In this process, the instrumental variable representation is obtained. Instrumental variable representation based on the current moment Energy management strategy for the next moment Calculate the correlation mutual information loss of instrumental variables. Instrumental variable representation based on the current moment Electricity consumption forecast results for the next moment Given a sequence of confusing variables at a historical moment Energy consumption management strategy sequence at historical moments Energy management strategy for the next moment Current electricity consumption results As a condition, calculate the exclusive mutual information loss of the instrumental variables. ; Representation of the confusing variable based on the current moment Energy management strategy for the next moment Calculate the mutual information loss of the correlation between the confounding variable and the intervention variable. ; Representation of the confusing variable based on the current moment The electricity consumption results at the next moment Calculate the correlation mutual information loss between the confounding variable and the outcome variable. Instrumental variable representation based on the current moment Sequence of confusing variables at all historical moments Calculate the mutual information loss of independence between instrumental variables and confounding variables. The sum of all the above losses is taken as the mutual information loss for overall representation learning. Its functional form is: , Get the parameters of this module .

[0047] Additionally, it should be noted that step one of this embodiment mainly includes the optimization process of decoupling representation learning. Given all electricity consumption conditions, energy management strategies, and historical information on electricity consumption, the time-series basic representation is first learned, and the ultimate goal is to decouple the representations of confounding variables and instrumental variables.

[0048] Step 2: Characterize the confounding variables at historical moments Energy management strategies for historical moments Instrumental variable representation at the current moment Energy management strategy for the next moment The data is input into the counterfactual regression module to predict the electricity consumption at the next moment. .exist Figure 3 The upper left section shows the processing flow of the counterfactual regression module.

[0049] It should be noted that the counterfactual regression module in this embodiment includes an outcome prediction network. Corresponding network parameters An energy management strategy prediction network Corresponding network parameters In the counterfactual regression module, the first... During each training round, the parameters of the long sequence modeling module, updated in the previous training round, are obtained. and The sequence of confusing variables at historical moments. Energy consumption management strategy sequence at historical moments Energy management strategy for the next moment Input the result prediction network In this process, the predicted electricity consumption for the next time moment is obtained. ; Representing the sequence of confusing variables at historical moments Energy consumption management strategy sequence at historical moments Instrumental variable representation at the current moment Input the energy management strategy prediction network In this process, the prediction results of the energy consumption management strategy are obtained. Based on the electricity consumption forecast results for the next time point The actual electricity consumption at the next moment Calculate the prediction loss of the resulting variable. Prediction results based on energy consumption management strategies Electricity consumption forecast results for the next moment The actual electricity consumption at the next moment Calculate the adversarial generalized moment conditional loss ; mutual information loss learned from the overall representation The prediction loss of the outcome variable Combating generalized moment conditional loss The weighted summation is used as the total loss of the algorithm. Its functional form is: , Optimize the parameters of the overall algorithm using gradient descent. , , The training is iterated continuously until the long sequence modeling module and the counterfactual regression module converge, and finally the trained long sequence modeling module and the counterfactual regression module are obtained.

[0050] gather Figure 4The algorithm flowchart needs further explanation. In each round of model training, step one is executed first, followed by step two, to obtain the total loss of the algorithm. Then backpropagation is performed to optimize the parameters. Then, training is repeated alternately. The representation learning network in step one mainly utilizes log-likelihood loss. Optimize the approximation of the variational distribution. Then, perform alternating training. Adversarial generalized moment conditional loss in step two of the round The aim is to utilize confounding variables and instrumental variables Predictive intervention variables In the worst-case scenario, it is possible to predict potential future outcomes as accurately as possible. .

[0051] Furthermore, in this embodiment, the aforementioned basic temporal representation learning network and the result prediction network All use transformers; the confusion variable characterization network The instrumental variable representation network and the energy consumption management strategy prediction network. All of them employ a multilayer perceptron (MLP). Further details will not be provided.

[0052] Example 2: This embodiment provides a causal effect estimation system for power consumption management strategies under unobserved confusion, which is used to implement the causal effect estimation method as described in Embodiment 1.

[0053] Example 3: like Figure 5 As shown, this embodiment provides an electronic device, which may include: at least one processor, at least one network interface, a user interface, a memory, and at least one communication bus.

[0054] The communication bus can be used to enable communication between the various components mentioned above.

[0055] The user interface may include buttons, and optional user interfaces may also include standard wired interfaces and wireless interfaces.

[0056] The network interface may include, but is not limited to, Bluetooth modules, NFC modules, Wi-Fi modules, etc.

[0057] The processor may include one or more processing cores. It connects various parts of the electronic device via various interfaces and lines, executing instructions, programs, code sets, or instruction sets stored in memory, and accessing data stored in memory to perform various functions and process data. Optionally, the processor can be implemented using at least one hardware form of DSP, FPGA, or PLA. The processor may integrate one or more of the following: CPU, GPU, and modem. The CPU primarily handles the operating system, user interface, and applications; the GPU is responsible for rendering and drawing the content required for display; and the modem handles wireless communication. It is understood that the modem may also be implemented as a separate chip without being integrated into the processor.

[0058] The memory may include RAM or ROM. Optionally, the memory may include a non-transitory computer-readable medium. The memory can be used to store instructions, programs, code, code sets, or instruction sets. The memory may include a program storage area and a data storage area, wherein the program storage area may store instructions for implementing an operating system, instructions for at least one function (such as touch function, sound playback function, image playback function, etc.), instructions for implementing the above-described method embodiments, etc.; the data storage area may store data involved in the above-described method embodiments, etc. Optionally, the memory may also be at least one storage device located remotely from the aforementioned processor. The memory, as a computer storage medium, may include an operating system, a network communication module, a user interface module, and an estimation application program. The processor can be used to call the estimation application program stored in the memory and execute the steps of the causal effect estimation method mentioned in the foregoing embodiments.

[0059] Example 4: This embodiment provides a computer-readable storage medium storing instructions that, when executed on a computer or processor, cause the computer or processor to perform the above-described instructions. Figures 1-4 One or more steps in the illustrated embodiment. If the constituent modules of the above-described electronic device are implemented as software functional units and sold or used as independent products, they can be stored in the computer-readable storage medium.

[0060] In the above embodiments, implementation can be achieved, in whole or in part, through software, hardware, firmware, or any combination thereof. When implemented in software, it can be implemented, in whole or in part, as a computer program product. The computer program product includes one or more computer instructions. When the computer program instructions are loaded and executed on a computer, all or part of the processes or functions described in the embodiments of this specification are generated. The computer can be a general-purpose computer, a special-purpose computer, a computer network, or other programmable device. The computer instructions can be stored in or transmitted through a computer-readable storage medium. The computer instructions can be transmitted from one website, computer, server, or data center to another website, computer, server, or data center via wired (e.g., coaxial cable, fiber optic, Digital Subscriber Line (DSL)) or wireless (e.g., infrared, wireless, microwave, etc.) means. The computer-readable storage medium can be any available medium accessible to a computer or a data storage device such as a server or data center that integrates one or more available media. The available media may be magnetic media (e.g., floppy disks, hard disks, magnetic tapes), optical media (e.g., Digital Versatile Discs (DVDs)), or semiconductor media (e.g., Solid State Disks (SSDs)).

[0061] Those skilled in the art will understand that all or part of the processes in the method of Embodiment 1 described above can be implemented by a computer program instructing related hardware. This program can be stored in a computer-readable storage medium, and when executed, it can include the processes of the embodiments of the methods described above. The aforementioned storage medium includes various media capable of storing program code, such as ROM, RAM, magnetic disks, or optical disks. Unless otherwise specified, the technical features of this embodiment and the implementation scheme can be combined arbitrarily.

[0062] Example 5: To verify the effectiveness of the causal effect estimation method for power consumption management strategy under unobserved confusion described in this specification, this embodiment is validated on a publicly available ICU patient dataset (MIMIC-III), a cryptocurrency dataset, a synthetic dataset simulating tumor growth, and a simulation dataset.

[0063] The ICU patient dataset (MIMIC-III) contains de-identified health data from patients admitted to the intensive care unit of a large tertiary hospital, derived from existing literature: Johnson AE, Pollard TJ, Shen L, et al. MIMIC-III, a freely accessible critical care database, Sci[J]. Data,2016, 3(1): 1-9. Following the data processing methods in MIMIC-extract, this embodiment obtained 69 covariates related to patient vital signs. Vasopressors and mechanical ventilation are two treatment measures, while blood pressure is the outcome variable of interest. MIMIC-extract is derived from existing literature: Wang S, McDermott MBA, Chauhan G, et al. Mimic-extract: A data extraction, preprocessing, and representation pipeline for mimic-iii[C]. Proceedings of the ACM conference on health, inference, and learning. 2020: 222-235.

[0064] The Cryptocurrency dataset collects real-time time-series data for Bitcoin and Ethereum from July 1, 2023 to October 29, 2024. Covariates include the opening price, lowest price, highest price, closing price, and trading volume for each day. The intervention variable is individual holdings, and the outcome variable is the rate of return.

[0065] The synthetic dataset simulating tumor growth was derived from existing literature: Geng C, Paganetti H, Grassberger C. Prediction of treatment response for combined chemotherapy and radiation therapy for non-small cell lung cancer patients using a bio-mathematical model[J]. Scientific reports, 2017, 7(1): 13542. This embodiment selected patient type and two static features as covariates. Treatment measures included chemotherapy and radiotherapy. The outcome measure of interest was tumor volume. Considering the scope of the data, this embodiment... and Normalization was performed.

[0066] This embodiment also generates a simulation dataset. For each time step... This embodiment generates 3D instrumental variables. And 7-dimensional confounding variables Each dimension follows a uniform distribution Generate 3D unobserved confusion. And each dimension follows a normal distribution. Obtained through random sampling. Update the time series using the following formula: , Intervention in the training and validation sets The following was generated: , , , Among them, coefficient It is randomly generated; this embodiment initializes it. and To evaluate the method's ability to predict counterfactual events, sequences in the test set were used. It is generated completely randomly. This example further generates the following result variables: .

[0067] Baseline methods include the TSD model, CT model, ACTIN model, ORL model, and DLTMLE model. The TSD model is derived from the existing literature: Bica I, Alaa A, Van Der Schaar M. Time series deconfounder: Estimating treatment effects over time in the presence of hidden confounders[C]. International conference on machine learning. PMLR, 2020: 884-895. The CT model is derived from the existing literature: Melnychuk V, Frauen D, Feuerriegel S. Causal transformer for estimating counterfactual outcomes[C]. Internationalconference on machine learning. PMLR, 2022: 15293-15329. The ACTIN model is derived from the existing literature: Wang X, Lyu S, Yang L, et al. A dual-module framework for counterfactual estimation over time[C]. Forty-first International Conference on Machine Learning. 2024. The ORL model is derived from existing technical literature: Tran A, Bibaut A, Kallus N. Inferring the long-term causal effects of long-term treatments from short-term experiments[C]. Proceedings of the 41st International Conference on Machine Learning. 2024: 48565-48577.The Deep LTMLE model is derived from existing technical literature: Shirakawa T, Li Y, Wu Y, et al. Longitudinal targeted minimum loss-based estimation with temporal-difference heterogeneous transformer[C]. Proceedings of the 41st International Conference on Machine Learning. 2024: 45097-45113.

[0068] The experimental results for single-step prediction in this embodiment are summarized in Table 1, using the mean squared error (MSE) metric. To ensure the robustness and reliability of the experimental results, each experiment was independently repeated 5 times, each time using a different random seed. All results are presented in the form of (mean ± standard deviation). As can be seen from Table 1, the method proposed in this invention significantly outperforms state-of-the-art baseline methods on all datasets.

[0069] Table 1. Results of causal effect estimation on multiple datasets:

[0070] It should be noted that the lower the value, the better the performance; the best performance in Table 1 is indicated by bold.

[0071] This embodiment is also extended to decision-making for sequential interventions. CT and ACTIN can be extended to predict the future. The result of the step, that is Their main idea is to additionally predict the next moment. The value is used in this way, and the estimated value is used in subsequent steps. and Instead of continuing to predict the observations required for the cycle, this method can be used. While the method of this invention can be easily extended in the manner described above—that is, obtaining the final result by iteratively repeating the estimation of a single time step—this embodiment attempts to directly predict the future outcome of sequential therapy. This embodiment naturally considers applying this model to downstream tasks, namely, making decisions about the optimal sequential intervention in the next time period. This embodiment has experimented with multi-step predictive decision-making and will... Set to 5. To obtain oracle results for all possible intervention sequences, this embodiment uses a simulated dataset for traversal generation. Figure 6 The report described CT, ACTIN, and the optimal intervention sequence determined by this invention. The corresponding actual result The results are compared with those from Oracle. It can be seen that the method proposed in this invention performs well and achieves performance close to that of Oracle.

[0072] Based on the above, this embodiment verifies the effectiveness of the causal effect estimation method for power consumption management strategy under unobserved confusion as described in this specification.

[0073] It should be noted that, for the sake of simplicity, the foregoing method embodiments are all described as a series of actions. However, those skilled in the art should understand that the present invention is not limited to the described order of actions, because according to the present invention, some steps can be performed in other orders or simultaneously. Furthermore, those skilled in the art should also understand that the embodiments described in the specification are preferred embodiments, and the actions and modules involved are not necessarily essential to the present invention.

[0074] In the above embodiments, the descriptions of each embodiment have different focuses. For parts not described in detail in a certain embodiment, please refer to the relevant descriptions in other embodiments.

[0075] The above description is merely an exemplary embodiment of the present invention and should not be construed as limiting the scope of the invention. Any equivalent changes and modifications made in accordance with the teachings of this invention are still within the scope of this invention. Those skilled in the art will readily conceive of embodiments of the invention upon considering the specification and practicing the disclosure herein. This invention is intended to cover any variations, uses, or adaptations of the invention that follow the general principles of the invention and include common knowledge or customary techniques in the art not described herein. The specification and embodiments are to be considered exemplary only, and the scope and spirit of the invention are defined by the claims.

Claims

1. A method for estimating the causal effect of energy control strategies under unobserved confounding for electricity, characterized in that, The method comprises the steps of: S1, acquire a training data set, the training data set comprising power consumption history information at a historical time , a use energy management and control strategy sequence , and a power consumption sequence ; S2, based on the training data set, the long sequence modeling module and the counterfactual regression module are trained end to end, and the joint iterative training comprises: S21. The electricity consumption history information at the aforementioned historical time point... Energy consumption management strategy sequence and electricity consumption sequence The input is fed into the long sequence modeling module to learn the representation of the confusion variables. and instrumental variable representation The confusion variable characterizes The instrumental variables are used to characterize incompletely observed variables that simultaneously affect energy management strategies and electricity consumption. Used to characterize variables that are relevant to energy management strategies and are not subject to unobserved confounding; S22, representing the confusion variable , the energy use control strategy sequence , the tool variable representation , and the next time energy use control strategy input into the counterfactual regression module, and the next time electricity consumption is predicted ; S3, input the to-be-estimated sequence data into the trained long sequence modeling module and the counterfactual regression module, execute steps S21 and S22, predict the power consumption under the target energy management and control strategy, and complete the causal effect estimation.

2. The method of claim 1, wherein the method is characterized by, The long sequence modeling module comprises a basic timing representation learning network , a confounding variable representation network , and an instrumental variable representation network , and the step S21 specifically comprises: combining the power consumption history information of the historical moment , the energy management and control strategy sequence , and the power consumption sequence into a historical characterization sequence ; Utilizing the base timing representation learning network on the history representation sequence processing, obtaining a base representation ; characterizing the base table inputting the obfuscated variable representation network obtaining an obfuscated variable representation ; characterizing the base table inputting the tool variable characterizing network , obtaining tool variable characterization .

3. The method of claim 2, wherein the method further comprises: The counterfactual regression module includes a result prediction network and a controllable policy prediction network Step S22 is specifically: characterizing the confusion variable , the energy consumption control strategy sequence and the predicted next time energy consumption control strategy inputting the result prediction network , obtaining the electricity consumption prediction result ; characterizing the confusion variable , the energy management strategy sequence and the tool variable inputting the strategy prediction network , obtaining a prediction result of the energy management strategy .

4. The method of claim 3, wherein, In the joint iterative training of step S2, the model parameters are updated by optimizing a total loss function of the algorithm, whose expression is: ​​ ， In the formula, is the prediction result of the electricity consumption at the next moment and the true result of the electricity consumption the prediction loss of the result variable calculated, is the mutual information loss of the overall representation learning, is the loss of the adversarial generalized moment condition, and is a hyperparameter.

5. The method of claim 4, wherein, In the training process of the long sequence modeling module, mutual information loss of overall representation learning is calculated in a functional form: ， wherein, is the correlation mutual information loss of the instrumental variable calculated based on the and the use energy control strategy of the next moment is the correlation mutual information loss of the instrumental variable calculated based on the and the use energy control strategy of the next moment is the correlation mutual information loss of the instrumental variable calculated based on the and the use energy control strategy of the next moment is the correlation mutual information loss of the instrumental variable calculated based on the and the use energy control strategy of the next moment is the correlation mutual information loss of the instrumental variable calculated based on the and the use energy control strategy of the next moment is the correlation mutual information loss of the instrumental variable calculated based on the and the use energy control strategy of the next moment is the correlation mutual information loss of the instrumental variable calculated based on the and the use energy control strategy of the next moment is the correlation mutual information loss of the instrumental variable calculated based on the and the use energy control strategy of the next moment is the correlation mutual information loss of the instrumental variable calculated based on the and the use energy control strategy of the next moment is the correlation mutual information loss of the instrumental variable calculated based on the 6. The method of claim 5, wherein the method further comprises: The function form of the adversarial GMM loss is: ， In the formula, is the number of samples, is the length of the time series, is the output of the energy management strategy prediction network, is the output of the result prediction network, is the real power consumption.

7. The method according to claim 6, wherein the method is used for estimating the causal effect of an unobserved confounding under a power energy management and control strategy. The base timing table representation learning network The result prediction network Both employ a transformer model; The obfuscated variable representation network The tool variable representation network And the energy usage control policy prediction network Each employs a multi-layer perceptron model.

8. A causal effect estimation system for power consumption management strategies under unobserved confusion, characterized in that, A computer program product for implementing the method according to any one of claims 1 to 7.

9. A computer device comprising a memory, a processor and a computer program, characterized in that The computer program is executed by a processor to implement the method according to any one of claims 1 to 7.

10. A computer-readable storage medium having stored thereon a computer program, characterized in that, The computer program is executed by a processor to implement the method according to any one of claims 1 to 7. The computer program is executed by a processor to implement the method according to any one of claims 1 to 7.

Images